This work developed a novel dataset generation framework based on Adaptive Augmentation and Synthetic Image Generation (AAg-SiG), which improves the image quality in the original dataset and generates synthetic images to balance the input dataset. The proposed framework works on the class-based feature adaptation method to fix anomalies in the images, like brightness, colour variations, blur using Gaussian Blur, CLAHE and UnsharpMark methods. The Oriented FAST and Rotated BRIEF (ORB) technique was used for geometric augmentation by using controlled affine transformation and key point detection. Moreover, the class mean (soft colour alignment) was used to fix the RGB distribution of the image, which preserved the specific features of all the images in the respective dataset class. Additionally, the interquartile range (IQR) was used to generate the class-specific synthetic image by calculating RGB channel features. Besides, the final stage was the validation of the generated image; an input class-based pre-fixed quality metrics were used to compare the augmented and generated image for authentication. Furthermore, the evaluated ORB heatmaps results justify the effectiveness of the AAG-SiG framework, as the important features (i.e. edges and feature density) were presented in the generated images. All in all, the AAg-SiG can be a useful dataset augmentation and generation tool to integrate with real-world applications of AI-models in the food process industry for strengthening the intelligent quality assurance (QA) and quality control (QC) systems.